Resource Management, Pricing, and Congestion Control for High-Speed Networks

Tamer Basar

University of Illinois at Urbana-Champaign


Communication networks, such as the Internet or ATM networks, can be viewed as largescale systems with multiple levels of decision making, involving service providers (at the higher level) and individual users (at the lower level). The decisions that each user is faced with, which are generally guided by an appropriate performance index, are (i) at what (flow) rate to inject packets into the network, and (ii) how to adjust these rates in response to (delayed) congestion information received from the network. Occasionally, users are also faced with the task of making decisions on routing, again based on information received from the network. The decisions the network service providers are faced with, on the other hand, are what to charge for the offered resources (such as bandwidth on each link), on what basis to admit users to the network, and whether or not to add additional capacity to any of the links--all driven by revenue maximization. Modelling and analyzing such systems, as well as architecting such networks and constructing routing and congestion control policies present challenges of mathematical, engineering and economic nature. This talk will address a number of issues that arise in this context, and present mathematical models that capture the decision making processes at the higher as well as the lower levels. The model for the former is game-theoretic, with the service provider being the revenue-maximizing price-setter and the users being utility maximizing price-takers, with the decisions made on a slower time scale. For the lower level, on the other hand, the model is a large-scale nonlinear control system with uncertain delays, which operates on a relatively faster time scale and with decentralization. For both models, some concrete results will be presented on issues such as optimum pricing, admission control, and capacity expansion at the higher level, and stability of decentralized and distributed rate control algoritms at the lower level (such as TCP and AQM schemes). The talk will conclude with a discussion of some open problems in this area.

[Tamer Basar]
[University of Illinois at Urbana-Champaign]